265 lines
7.4 KiB
C++
265 lines
7.4 KiB
C++
/* ----------------------------------------------------------------------------
|
|
|
|
* GTSAM Copyright 2010, Georgia Tech Research Corporation,
|
|
* Atlanta, Georgia 30332-0415
|
|
* All Rights Reserved
|
|
* Authors: Frank Dellaert, et al. (see THANKS for the full author list)
|
|
|
|
* See LICENSE for the license information
|
|
|
|
* -------------------------------------------------------------------------- */
|
|
|
|
/**
|
|
* @file testTOAFactor.cpp
|
|
* @brief Unit tests for "Time of Arrival" factor
|
|
* @author Frank Dellaert
|
|
* @author Jay Chakravarty
|
|
* @date December 2014
|
|
*/
|
|
|
|
#include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h>
|
|
#include <gtsam_unstable/nonlinear/ExpressionFactor.h>
|
|
#include <gtsam/geometry/Point3.h>
|
|
#include <cmath>
|
|
|
|
namespace gtsam {
|
|
|
|
/// A space-time event
|
|
class Event {
|
|
|
|
double time_; ///< Time event was generated
|
|
Point3 location_; ///< Location at time event was generated
|
|
|
|
public:
|
|
|
|
/// Speed of sound
|
|
static const double Speed;
|
|
|
|
/// Default Constructor
|
|
Event() :
|
|
time_(0) {
|
|
}
|
|
|
|
/// Constructor from time and location
|
|
Event(double t, const Point3& p) :
|
|
time_(t), location_(p) {
|
|
}
|
|
|
|
/// Constructor with doubles
|
|
Event(double t, double x, double y, double z) :
|
|
time_(t), location_(x, y, z) {
|
|
}
|
|
|
|
/** print with optional string */
|
|
void print(const std::string& s = "") const {
|
|
std::cout << s << ", time = " << time_ << std::endl;
|
|
location_.print("location");
|
|
}
|
|
|
|
/** equals with an tolerance */
|
|
bool equals(const Event& other, double tol = 1e-9) const {
|
|
return std::abs(time_ - other.time_) < tol
|
|
&& location_.equals(other.location_, tol);
|
|
}
|
|
|
|
/// Manifold stuff:
|
|
|
|
size_t dim() const {
|
|
return 4;
|
|
}
|
|
static size_t Dim() {
|
|
return 4;
|
|
}
|
|
|
|
/// Updates a with tangent space delta
|
|
inline Event retract(const Vector4& v) const {
|
|
return Event(time_ + v[0], location_.retract(v.tail(3)));
|
|
}
|
|
|
|
/// Returns inverse retraction
|
|
inline Vector4 localCoordinates(const Event& q) const {
|
|
return Vector4::Zero(); // TODO
|
|
}
|
|
|
|
/// Time of arrival to given microphone
|
|
double toa(const Point3& microphone, //
|
|
OptionalJacobian<1, 4> H1 = boost::none, //
|
|
OptionalJacobian<1, 3> H2 = boost::none) const {
|
|
Matrix13 D1, D2;
|
|
double distance = location_.distance(microphone, D1, D2);
|
|
if (H1) {
|
|
// derivative of toa with respect to event
|
|
*H1 << 1, D1 / Speed;
|
|
}
|
|
if (H2) {
|
|
// derivative of toa with respect to microphone location
|
|
*H2 << D2 / Speed;
|
|
}
|
|
return time_ + distance / Speed;
|
|
}
|
|
};
|
|
|
|
const double Event::Speed = 330;
|
|
|
|
// Define GTSAM traits
|
|
namespace traits {
|
|
|
|
template<>
|
|
struct GTSAM_EXPORT dimension<Event> : public boost::integral_constant<int, 4> {
|
|
};
|
|
|
|
}
|
|
|
|
/// A "Time of Arrival" factor
|
|
class TOAFactor: public ExpressionFactor<double> {
|
|
|
|
typedef Expression<double> double_;
|
|
|
|
public:
|
|
|
|
/**
|
|
* Constructor
|
|
* @param some expression yielding an event
|
|
* @param microphone_ expression yielding a microphone location
|
|
* @param toaMeasurement time of arrival at microphone
|
|
* @param model noise model
|
|
*/
|
|
TOAFactor(const Expression<Event>& event_,
|
|
const Expression<Point3>& microphone_, double toaMeasurement,
|
|
const SharedNoiseModel& model) :
|
|
ExpressionFactor<double>(model, toaMeasurement,
|
|
double_(&Event::toa, event_, microphone_)) {
|
|
}
|
|
|
|
};
|
|
|
|
} //\ namespace gtsam
|
|
|
|
#include <gtsam/nonlinear/NonlinearFactorGraph.h>
|
|
#include <gtsam/base/numericalDerivative.h>
|
|
#include <CppUnitLite/TestHarness.h>
|
|
#include <boost/bind.hpp>
|
|
|
|
using namespace std;
|
|
using namespace gtsam;
|
|
|
|
// Create a noise model for the TOA error
|
|
//static const double ms = 1e-3;
|
|
static const double cm = 1e-2;
|
|
typedef Eigen::Matrix<double, 1, 1> Vector1;
|
|
static SharedNoiseModel model(noiseModel::Unit::Create(1));
|
|
|
|
//*****************************************************************************
|
|
TEST( Event, Constructor ) {
|
|
const double t = 0;
|
|
Event actual(t, 201.5 * cm, 201.5 * cm, (212 - 45) * cm);
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST( Event, Toa1 ) {
|
|
Point3 microphone;
|
|
Event event(0, 1, 0, 0);
|
|
double expected = 1 / Event::Speed;
|
|
EXPECT_DOUBLES_EQUAL(expected, event.toa(microphone), 1e-9);
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST( Event, Toa2 ) {
|
|
Point3 microphone;
|
|
double timeOfEvent = 25;
|
|
Event event(timeOfEvent, 1, 0, 0);
|
|
double expectedTOA = timeOfEvent + 1 / Event::Speed;
|
|
EXPECT_DOUBLES_EQUAL(expectedTOA, event.toa(microphone), 1e-9);
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST( Event, Expression ) {
|
|
Key key = 12;
|
|
Expression<Event> event_(key);
|
|
Point3 microphone;
|
|
Expression<Point3> knownMicrophone_(microphone); // constant expression
|
|
Expression<double> expression(&Event::toa, event_, knownMicrophone_);
|
|
|
|
// double timeOfEvent = 25;
|
|
// Event event12(timeOfEvent, 1, 0, 0);
|
|
// Values values;
|
|
// values.insert(key,event12);
|
|
// double expectedTOA = timeOfEvent + 1 / Event::Speed;
|
|
// EXPECT_DOUBLES_EQUAL(expectedTOA, expression.value(values), 1e-9);
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST(Event, Retract) {
|
|
Event event, expected(1, 2, 3, 4);
|
|
Vector4 v;
|
|
v << 1, 2, 3, 4;
|
|
EXPECT(assert_equal(expected, event.retract(v)));
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST( TOAFactor, Construct ) {
|
|
Key key = 12;
|
|
Expression<Event> event_(key);
|
|
Point3 microphone;
|
|
Expression<Point3> knownMicrophone_(microphone); // constant expression
|
|
double measurement = 7;
|
|
TOAFactor factor(event_, knownMicrophone_, measurement, model);
|
|
}
|
|
|
|
//*****************************************************************************
|
|
TEST( TOAFactor, WholeEnchilada ) {
|
|
|
|
// Create microphones
|
|
vector<Point3> microphones;
|
|
microphones.push_back(Point3(0, 0, 0));
|
|
microphones.push_back(Point3(403 * cm, 0, 0));
|
|
microphones.push_back(Point3(403 * cm, 403 * cm, 0));
|
|
microphones.push_back(Point3(0, 403 * cm, 0));
|
|
EXPECT_LONGS_EQUAL(4, microphones.size());
|
|
|
|
// Create a ground truth point
|
|
const double timeOfEvent = 0;
|
|
Event groundTruthEvent(timeOfEvent, 201.5 * cm, 201.5 * cm, (212 - 45) * cm);
|
|
|
|
// Simulate measurements
|
|
vector<double> measurements(4);
|
|
for (size_t i = 0; i < 4; i++)
|
|
measurements[i] = groundTruthEvent.toa(microphones[i]);
|
|
|
|
// Now, estimate using non-linear optimization
|
|
NonlinearFactorGraph graph;
|
|
Key key = 12;
|
|
Expression<Event> event_(key);
|
|
for (size_t i = 0; i < 4; i++) {
|
|
Expression<Point3> knownMicrophone_(microphones[i]); // constant expression
|
|
graph.add(TOAFactor(event_, knownMicrophone_, measurements[i], model));
|
|
}
|
|
|
|
/// Print the graph
|
|
GTSAM_PRINT(graph);
|
|
|
|
// Create initial estimate
|
|
Values initialEstimate;
|
|
Event estimatedEvent(timeOfEvent + 0.1, 200 * cm, 150 * cm, 50 * cm);
|
|
initialEstimate.insert(key, estimatedEvent);
|
|
|
|
// Print
|
|
initialEstimate.print("Initial Estimate:\n");
|
|
|
|
// Optimize using Levenberg-Marquardt optimization.
|
|
LevenbergMarquardtParams params;
|
|
params.setVerbosity("ERROR");
|
|
LevenbergMarquardtOptimizer optimizer(graph, initialEstimate);
|
|
Values result = optimizer.optimize();
|
|
result.print("Final Result:\n");
|
|
EXPECT(assert_equal(groundTruthEvent, result.at<Event>(key)));
|
|
}
|
|
|
|
//*****************************************************************************
|
|
int main() {
|
|
TestResult tr;
|
|
return TestRegistry::runAllTests(tr);
|
|
}
|
|
//*****************************************************************************
|
|
|